Perceptual Learning, Roving and the Unsupervised Bias
description
Transcript of Perceptual Learning, Roving and the Unsupervised Bias
![Page 1: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/1.jpg)
Perceptual Learning, Roving and the Unsupervised Bias
By Aaron Clarke, Henning Sprekeler, Wolfram Gerstner and Michael Herzog
Brain Mind InstituteÉcole Polytechnique Fédérale De Lausanne
Switzerland
![Page 2: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/2.jpg)
Talk Outline
• Perceptual Learning & Roving• The Unsupervised Bias• Critical Experiment
![Page 3: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/3.jpg)
Perceptual Learning
![Page 4: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/4.jpg)
Perceptual Learning
0 5 10 15 20 25 30 35 40 45 500
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Block Number
d'
![Page 5: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/5.jpg)
Talk Outline
• Perceptual Learning & Roving• The Unsupervised Bias• Critical Experiment
![Page 6: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/6.jpg)
Roving
1200”1200”
Learning Task 1
0 5 10 15 20 25 30 35 40 45 500
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Block Number
d'
![Page 7: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/7.jpg)
Roving
1200”1200” 1800” 1800”
Learning Task 1 Learning Task 2
![Page 8: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/8.jpg)
![Page 9: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/9.jpg)
![Page 10: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/10.jpg)
![Page 11: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/11.jpg)
![Page 12: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/12.jpg)
![Page 13: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/13.jpg)
Roving
0 5 10 15 20 250.5
1
1.5
2
2.5
3
3.5
4
Block Number
d'
Non-Roved
0 5 10 15 20 250.5
1
1.5
2
2.5
3
3.5
4
Block Number
d'
Roved
1200"1800"
Adapted from Tartaglia, Bamert, Mast & H. Herzog (2009)
![Page 14: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/14.jpg)
Hypotheses
• Roving may disrupt memory-trace-buildup for the roved stimuli (Yu et al., 2004).
• Roving may diminish the stimuli’s predictability (Adini et al., 2004).
• Roving may prevent the participants from conceptually tagging each stimulus type in order to switch their attention to the appropriate perceptual template (Zhang et al., 2008).
![Page 15: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/15.jpg)
Roving
1200”1200” 1800” 1800”
Learning Task 1 Learning Task 2
![Page 16: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/16.jpg)
Hypotheses
• Roving may disrupt memory-trace-buildup for the roved stimuli (Yu et al., 2004).
• Roving may diminish the stimuli’s predictability (Adini et al., 2004).
• Roving may prevent the participants from conceptually tagging each stimulus type in order to switch their attention to the appropriate perceptual template (Zhang et al., 2008).
![Page 17: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/17.jpg)
Talk Outline
• Perceptual Learning & Roving• The Unsupervised Bias• Critical Experiment
![Page 18: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/18.jpg)
Talk Outline
• Perceptual Learning & Roving• The Unsupervised Bias• Critical Experiment
![Page 19: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/19.jpg)
Model Predictions
SupervisedUnsupervised Reward-Based
• No feedback • Trial by trial feedback
• Error feedback• Teacher signal
Output Desired Output
Error
Input
Output Desired Output
Error
Reward
• Feedback after many trials• Error feedback• Teacher signal
i
j
InputInput
Δwij = prei × eij Δwij = Cov(R,wij) + ‹R› ‹wij›Δwij = prei × postj
![Page 20: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/20.jpg)
Δwij = prei × eij
Model Predictions
Unsupervised Reward-Based
• No feedback
Input
Output Desired Output
Error
Reward
• Feedback after many trials• Error feedback• Teacher signal
i
j
Supervised
• Trial by trial feedback
• Error feedback• Teacher signal
Input
Output Desired Output
Error
Herzog & Fahle (1998)
Feedback improves performance.
Learning is possible without feedback
Δwij = Cov(R,wij) + ‹R› ‹wij›Δwij = prei × postj
![Page 21: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/21.jpg)
Reward-Based Learning
Δwij = Cov(R,wij) + ‹R› ‹wij›
weight change Covariation between reward weight change
Average reward
Averages of past trials Reward & current activations
![Page 22: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/22.jpg)
Reward-Based Learning
Δwij = Cov(R,wij) + ‹R› ‹wij›
weight change Covariation between reward weight change
Average reward
= 0
Averages of past trials Reward & current activations
![Page 23: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/23.jpg)
Reward-Based Learning
Δwij = Cov(R1+R2,wij) + ‹R1+R2› ‹wij›
weight change Covariation between reward weight change
Average reward
Averages of past trials
• Learning is impossible with two stimuli.
Reward & current activations
![Page 24: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/24.jpg)
Roving
0 5 10 15 20 250.5
1
1.5
2
2.5
3
3.5
4
Block Number
d'
Non-Roved
0 5 10 15 20 250.5
1
1.5
2
2.5
3
3.5
4
Block Number
d'
Roved
1200"1800"
Adapted from Tartaglia, Bamert, Mast & H. Herzog (2009)
![Page 25: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/25.jpg)
Talk Outline
• Perceptual Learning & Roving• The Unsupervised Bias• Critical Experiment
![Page 26: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/26.jpg)
Hypothesis
• Roving impairs perceptual learning when the average reward for the two learned stimuli differs significantly.– This kind of situation occurs when the two roved
tasks differ in their difficulty levels.
![Page 27: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/27.jpg)
Roving
1200”1200” 1800” 1800”
Learning Task 1 Learning Task 2
![Page 28: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/28.jpg)
Results
H0: Mean Hard Slopes = 0:t(7) = -1.115, p = 0.151
1200”
1800”
H0: Mean Easy Slopes = 0:t(7) = -0.222, p = 0.415
0 5 10 15 200
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Block Number
d'
EasyHard
![Page 29: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/29.jpg)
Results
0 5 10 15 200
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Block Number
d'
EasyHard
0 5 10 15 200
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Block Number
d'
H0: Mean Non-Roved Slopes = 0:t(7) = 2.144, p = 0.035
![Page 30: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/30.jpg)
Summary• There are three types of learning models: supervised,
unsupervised and reward-based.• Only reward-based learning withstands empirical
falsification, and it suffers from the unsupervised bias.• When roving two tasks, easy and hard, learning fails, as can
be shown mathematically. And that is why roving occurs empirically.
• A strange prediction from this is that roving a hard and a very easy task should deteriorate performance. Roving two hard tasks might make learning easier than roving a hard and an easy task, and this has actually been shown in other studies.
![Page 31: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/31.jpg)
Thank for your attention.
![Page 32: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/32.jpg)
When is Learning During Roving Successful?
Vs.
Vs.
150 ms 500 ms
Vs.
![Page 33: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/33.jpg)
Experiment• Used two stimuli: 1800” and
1200”.• Measured pre-training
thresholds for both stimuli in isolation.
• Trained subjects with fixed offsets (easy = 1.5 × pre-training threshold, hard = 0.9 × pre-training threshold).
• In 20 blocks of 80 trials.• Roved stimuli.
1800”
Easy
1200”
Hard
Easy
1200”
![Page 34: Perceptual Learning, Roving and the Unsupervised Bias](https://reader035.fdocuments.us/reader035/viewer/2022062305/56816302550346895dd37a16/html5/thumbnails/34.jpg)
Other Hypotheses
• Roving may interact with the participants’ initial performance levels where worse initial performers learn more than high initial performers.
• Roving might cause low-level interference between stimulus types (Tartaglia et al., 2009; Zhaoping, Herzog, & Dayan, 2003).